call meteor about nlg-eval HOT 7 CLOSED

Ah that's very helpful. It seems like you don't have much memory on your machine?
You can changing that '-Xmx2G' to something lower like '-Xmx1G' in meteor.py.

from nlg-eval.

It works now !
Thank you!
You've been a lifesaver.

from nlg-eval.

Thanks for reporting this. I can help you debug it.
Change that line from scores.append(float(dec(self.meteor_p.stdout.readline().strip()))) to these lines:

v = self.meteor_p.stdout.readline()
try:
    scores.append(float(dec(v.strip())))
except:
    print("Error handling value: {}".format(v))
    print("Decoded value: {}".format(dec(v.strip())))
    raise

and re-run the program. It should help us see what we're dealing with.

PS Is that your birthday in your GitHub username? If so you should consider changing it as many places irresponsibly use a birthday to authenticate people.

from nlg-eval.

Thanks your advice!

It print this
Error handling value: b''
Decoded value:

from nlg-eval.

I can't find a nice way to work in a fix and I can't replicate the issues that people get with Meteor.
For now can you add do this:

v = self.meteor_p.stdout.readline()
try:
    scores.append(float(dec(v.strip())))
except:
    sys.stderr.write("Error handling value: {}\n".format(v))
    sys.stderr.write("Decoded value: {}\n".format(dec(v.strip())))
    sys.stderr.write("eval_line: {}\n".format(eval_line))
    raise

and if the eval_line isn't helpful then also add:

sys.stderr.write("Error from Meteor:\n{}".format(self.meteor_p.stderr.read()))

before raise. That line is dangerous because it will hang if there is no stderr from Meteor. So if you find that after adding that line your program never ends, then stop you program and remove it.

from nlg-eval.

Thanks for your help !!

Error handling value: b'Could not reserve enough space for 2097152KB object heap\n'
Decoded value: Could not reserve enough space for 2097152KB object heap
eval_line: EVAL ||| Error occurred during initialization of VM

from nlg-eval.

Thanks for the feedback. I'll look into adding a warning to help others in the future.

from nlg-eval.